Conscious Knowledge Based Question Answering

Shirong Liu, Zixian Guo, Hongzhi Wang

Abstract: A model of consciousness proposed by neuroscientists in 1989 is called the theater model, which uses theater as an analogy to describe "what is consciousness". This paper simplifies the problem of question answering and uses theater model to simulate the question answering mechanism of the human brain. We extract a small amount of knowledge from Freebase and use it as agents' knowledge base. Then we build a multi-round question answering agent based on the theater model and Deep Q-learning. We train the two agents against each other, and finally analyze the training results. The results show that the question answering mechanism and training method designed can simulate the human dialogue scene well, and agents have different behavior when setting different rewarding parameters.